GLMS SEC has released a research report maintaining a "Recommended" rating for KNOWLEDGE ATLAS (ASX: 02513). The firm forecasts the company's revenue for 2026-2028 to reach 2.9 billion, 7.5 billion, and 20.6 billion yuan, respectively. Based on the closing price on June 17, 2026, the projected PS valuations for those years are 220x, 85x, and 31x.

The company has long adhered to a strategic positioning of "foundation model + platform + ecosystem." Its GLM series models have consistently ranked at the top of global coding leaderboards, maintaining a leading edge in intelligence. The GLM model itself constitutes a significant barrier to entry. As GLM deepens its penetration into meta-scenarios like coding and AI agents, paid token consumption has experienced exponential growth. Concurrently, through continuous optimization on the inference side, the company's gross margin level is expected to improve, with its commercialization progress showing strong performance.

Key Analyst Views

Recently, the company announced the launch of its latest flagship model, GLM-5.2. The model offers a long context window of up to 1 million tokens and continues to lead in long-range tasks. It will be open-sourced under the MIT license, which is expected to boost subsequent commercial application call volumes for the company's open platform and API services.

GLM-5.2's Coding Prowess and ZCODE 3.0 Integration

GLM-5.2 is the most capable open-source model from KNOWLEDGE ATLAS to date, supporting a genuinely usable 1M context while maintaining leadership in long-range tasks. Its coding performance is outstanding. According to the company's WeChat channel, GLM-5.2 achieved first place among globally available models on the Code Arena front-end development evaluation system, a platform with millions of users participating in blind tests. Another source reports that GLM-5.2 scored 81.43 points on the KingBench 3 leaderboard, ranking just behind Fable 5 (88.57) and Opus 4.8 (87.14).

Simultaneously, the company has updated to ZCODE 3.0. ZCode is a full-featured Agentic Development Environment (ADE) for Long Horizon Tasks, designed to enable AI agents to complete longer-span, multi-step development tasks in an end-to-end, stable, and controllable manner. ZCode 3.0 has undergone deep joint debugging and specialized optimization for GLM-5.2, focusing on enhancing the proprietary ZCode Agent capabilities. This results in tighter integration between the model, tools, and execution workflow, allowing complex development tasks to progress from planning through to verification within a single context.

Geopolitical Shifts Create Opportunity for Domestic Models

Anthropic's latest models, Fable 5 and Mythos 5, have been restricted from use by foreign entities. Anthropic stated that the U.S. government, citing national security authority, issued an export control directive requiring the suspension of all access to these AI models for any foreign principal, regardless of location, including Anthropic employees who are foreign citizens. The practical effect of this order is that Anthropic must immediately disable Fable 5 and Mythos 5 for all clients to ensure compliance, though access to all other Anthropic models remains unaffected.

The report suggests that KNOWLEDGE ATLAS's move to fully open GLM-5.2, launch its API, and open-source it under MIT at this juncture positions it well to capture the substitution demand from developers and enterprise clients for advanced domestic models.

Potential Risks Outlined

The report highlights risks including large language model commercialization falling short of expectations and intensifying market competition.